【 Early Access Policy - Free Access】
To My Followers:
As a free user myself — where every single Buzz is earned through hard work on bounties, i know how hard it is.
According to recent change of civitai creator program.
I will only use Early Access Program and will never make any LoRA permanently paid (membership system).
I firmly believe that free models are what drive the sustainability of this community.
You may wonder why there are few LoRAs release in 1 day / 1 week, it is like my working style, batch training and release.
I will occasionally release a few LoRAs every month.
I don't want the access program become pressure to my followers.
If you’d like to support my work, feel free to drop some Buzz via purchase / tips / donations.
If you're a free user, you can simply wait for the public release.
or feel free to message me — I can send you a Google Drive link or refund your Buzz (and you can keep the model).
Sake 🍶
A Japanese Filter, named in different era to present their style
The dataset contains some cover art, may generate some word art / watermark , you can avoid by adding neg prompt:
Negative prompt
(word:1.4), (alphabet:1.35), (characters:1.25) Description
Comments (11)
Bro amazing lora once again. Also thanks for thinking about the community with the change in the creator program.
Thank you 😍
My intent is to looking for a Japanese color tone filter for the image.
Personally, I prefer the HeiSei branch, which is modern, bright, high in brightness, and low in contrast, just fit the concept of Japanese color filtering.
@al123sdaw i love the effect this lora has on the image.
I searched the difference of the two models and i found out there are more.
For the next style if possible can you try, Taishō filter.
Oh yea... I completely forgot TaiSho, it is more suit instead of ShoWa tag.
thanks for noticing, i just change the version's name.
the style is close to end of MeiJi, early TaiSho, the era of Nijusseiki Denki Mokuroku, Demon Slayer.
I will reserved the version tag once i found a suitable style fit into it. www thank you so much.
@al123sdaw How to do get the image for the style
@ArtificalEverything for TaiSho , i updated the trigger words, thats the LoRA style trained on,
I prefer it bring plain, less contrast, watercolor , coarse, retro tone. I posted few posts using the trigger words. The example is not complete yet, I hope it can be more blurry and coarse texture.
all the above is the official textbook to use.
For the rest, you can explore any kind of way , those effect may surprising us.
@al123sdaw Thanks, but i think my question was confusing. what i was asking is that how did you gather the dataset images to train the lora? Sorry for confusion.
i just found i wrote 200 words lmaoo , i reply you in PM msg.
in short, self generated.
@al123sdaw Thanks this helps me to rethink how i should go about data collection. while i got experienced from training a concept lora, but the challenge is to do a style lora and this helps.
Also you said you self generate your own dataset for the loras does that mean you did the same for the ramune series?
@ArtificalEverything Ramune is based on the real gaming art, so those data are from Wiki (NieR reincreation) and official art book (FF / elden ring etc), data cleansing is another hell , spend many time writing scripts / manual capture from the book....
ahhh , i dont want to recall that memory ....